Talks

Encoding

We use dvgrab to rip the talks from the camera and ffmpeg to encode.

Mirrors

• taurine (90 Mbps) [It can probably push much more if it were on a gigabit link]
• citric-acid (250 Mbps)
• natural-flavours (200 Mbps)
• artificial-flavours (200 Mbps)
• perpugilliam (90 Mbps)

The list of mirrors can be found in /users/www/mirrors.txt; the web server selects a random mirror from this list at each file request. We now run snmpd so we have real-time access to network and cpu load data. We should therefore be able to write a more intelligent mirror selection script.

We also use mirror.cs to mirror on-campus.

QoS

We are currently looking at implementing QoS on the host end. IST's QoS plan is documented here: http://ist.uwaterloo.ca/~dkeenan/docs/QoS-plan-draft.html. IST recommends using "IP Precendence 1, PHB CS1, DSCP 8, CoS 1". The ports we want to limit are 80 (HTTP) and 6900-6999 (BitTorrent).

DSCP Information:

• http://tools.ietf.org/html/rfc2474
• http://www.cisco.com/warp/public/105/dscpvalues.html

CoS/802.1p Information:

• http://www.ieee802.org/1/pages/802.1D-2003.html

Faculty Computing Implementations:

• Science: http://www.freebsd.uwaterloo.ca/twiki/bin/view/Test/QoSProject

Our switch (HP ProCurve 2810) supports DSCP -> 802.1p mappings. Other ProCurve switches likely support this as well (need to check with dlgawley though).

The /etc/network/if-up.d/csc-mirror script (puppetized) adds the iptables rules we need.